summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorTom Stellard <thomas.stellard@amd.com>2014-02-24 18:47:58 +0000
committerTom Stellard <thomas.stellard@amd.com>2014-02-24 18:47:58 +0000
commit0a98bd5bb8a9b6cab765323c3a9a4000eaf13f26 (patch)
tree1af2e0d8ba6c84c557bb32797c9bf5ef8567597e
parentdbb832b83351cec97b025b61c26536ef50c3181c (diff)
downloadllvm-0a98bd5bb8a9b6cab765323c3a9a4000eaf13f26.tar.gz
llvm-0a98bd5bb8a9b6cab765323c3a9a4000eaf13f26.tar.bz2
llvm-0a98bd5bb8a9b6cab765323c3a9a4000eaf13f26.tar.xz
Merging r197503, r197505, r197520:
------------------------------------------------------------------------ r197520 | dexonsmith | 2013-12-17 12:28:21 -0800 (Tue, 17 Dec 2013) | 7 lines Assert that the last operand is actually EFLAGS This is another follow-up to r197503, after a post-commit review by Andy. <rdar://problem/15627766> ------------------------------------------------------------------------ r197505 | dexonsmith | 2013-12-17 08:20:37 -0800 (Tue, 17 Dec 2013) | 6 lines Setting the CPU in the new vaargs test Trying to fix buildbots after r197503 (test passes locally). <rdar://problem/15627766> ------------------------------------------------------------------------ r197503 | dexonsmith | 2013-12-17 07:54:45 -0800 (Tue, 17 Dec 2013) | 17 lines Revert "Revert "Mark vastart_save_xmm_regs as changing EFLAGS"" This reverts commit r197481, recommiting r197469 with an extra fix. The vastart_save_xmm_regs pseudo-instruction expands to a test and a branch, so it modifies EFLAGS. Mark it so, or else the scheduler might place it in the middle of another test+branch. This fixes a bug exposed by r192750, which changed the initial scheduler to source-order as part of enabling the MI Scheduler for X86. This re-commit changes the VASTART_SAVE_XMM_REGS custom inserter not to try to save %flags, and adds a test that catches the bad behavior of r197469. <rdar://problem/15627766> ------------------------------------------------------------------------ git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/release_34@202060 91177308-0d34-0410-b5e6-96231b3b80d8
-rw-r--r--lib/Target/X86/X86ISelLowering.cpp8
-rw-r--r--lib/Target/X86/X86InstrCompiler.td5
-rw-r--r--test/CodeGen/X86/vaargs.ll67
-rw-r--r--test/CodeGen/X86/vastart-defs-eflags.ll23
4 files changed, 100 insertions, 3 deletions
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 76eeb64650..06929dd188 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -15226,9 +15226,15 @@ X86TargetLowering::EmitVAStartSaveXMMRegsWithCustomInserter(
MBB->addSuccessor(EndMBB);
}
+ // Make sure the last operand is EFLAGS, which gets clobbered by the branch
+ // that was just emitted, but clearly shouldn't be "saved".
+ assert((MI->getNumOperands() <= 3 ||
+ !MI->getOperand(MI->getNumOperands() - 1).isReg() ||
+ MI->getOperand(MI->getNumOperands() - 1).getReg() == X86::EFLAGS)
+ && "Expected last argument to be EFLAGS");
unsigned MOVOpc = Subtarget->hasFp256() ? X86::VMOVAPSmr : X86::MOVAPSmr;
// In the XMM save block, save all the XMM argument registers.
- for (int i = 3, e = MI->getNumOperands(); i != e; ++i) {
+ for (int i = 3, e = MI->getNumOperands() - 1; i != e; ++i) {
int64_t Offset = (i - 3) * 16 + VarArgsFPOffset;
MachineMemOperand *MMO =
F->getMachineMemOperand(
diff --git a/lib/Target/X86/X86InstrCompiler.td b/lib/Target/X86/X86InstrCompiler.td
index 7d10b67bfe..5c8840823b 100644
--- a/lib/Target/X86/X86InstrCompiler.td
+++ b/lib/Target/X86/X86InstrCompiler.td
@@ -72,7 +72,7 @@ def ADJCALLSTACKUP64 : I<0, Pseudo, (outs), (ins i32imm:$amt1, i32imm:$amt2),
// x86-64 va_start lowering magic.
-let usesCustomInserter = 1 in {
+let usesCustomInserter = 1, Defs = [EFLAGS] in {
def VASTART_SAVE_XMM_REGS : I<0, Pseudo,
(outs),
(ins GR8:$al,
@@ -81,7 +81,8 @@ def VASTART_SAVE_XMM_REGS : I<0, Pseudo,
"#VASTART_SAVE_XMM_REGS $al, $regsavefi, $offset",
[(X86vastart_save_xmm_regs GR8:$al,
imm:$regsavefi,
- imm:$offset)]>;
+ imm:$offset),
+ (implicit EFLAGS)]>;
// The VAARG_64 pseudo-instruction takes the address of the va_list,
// and places the address of the next argument into a register.
diff --git a/test/CodeGen/X86/vaargs.ll b/test/CodeGen/X86/vaargs.ll
new file mode 100644
index 0000000000..ddeb7a336d
--- /dev/null
+++ b/test/CodeGen/X86/vaargs.ll
@@ -0,0 +1,67 @@
+; RUN: llc -mcpu=corei7-avx %s -o - | FileCheck %s --check-prefix=CHECK --check-prefix=NO-FLAGS
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx10.9.0"
+
+%struct.__va_list_tag = type { i32, i32, i8*, i8* }
+
+; Check that vastart gets the right thing.
+define i32 @sum(i32 %count, ...) nounwind optsize ssp uwtable {
+; CHECK: testb %al, %al
+; CHECK-NEXT: je
+; CHECK-NEXT: ## BB#{{[0-9]+}}:
+; CHECK-NEXT: vmovaps %xmm0, 48(%rsp)
+; CHECK-NEXT: vmovaps %xmm1, 64(%rsp)
+; CHECK-NEXT: vmovaps %xmm2, 80(%rsp)
+; CHECK-NEXT: vmovaps %xmm3, 96(%rsp)
+; CHECK-NEXT: vmovaps %xmm4, 112(%rsp)
+; CHECK-NEXT: vmovaps %xmm5, 128(%rsp)
+; CHECK-NEXT: vmovaps %xmm6, 144(%rsp)
+; CHECK-NEXT: vmovaps %xmm7, 160(%rsp)
+
+; Check that [EFLAGS] hasn't been pulled in.
+; NO-FLAGS-NOT: %flags
+
+ %ap = alloca [1 x %struct.__va_list_tag], align 16
+ %1 = bitcast [1 x %struct.__va_list_tag]* %ap to i8*
+ call void @llvm.va_start(i8* %1)
+ %2 = icmp sgt i32 %count, 0
+ br i1 %2, label %.lr.ph, label %._crit_edge
+
+.lr.ph: ; preds = %0
+ %3 = getelementptr inbounds [1 x %struct.__va_list_tag]* %ap, i64 0, i64 0, i32 0
+ %4 = getelementptr inbounds [1 x %struct.__va_list_tag]* %ap, i64 0, i64 0, i32 2
+ %.pre = load i32* %3, align 16
+ br label %5
+
+; <label>:5 ; preds = %.lr.ph, %13
+ %6 = phi i32 [ %.pre, %.lr.ph ], [ %14, %13 ]
+ %.01 = phi i32 [ %count, %.lr.ph ], [ %15, %13 ]
+ %7 = icmp ult i32 %6, 41
+ br i1 %7, label %8, label %10
+
+; <label>:8 ; preds = %5
+ %9 = add i32 %6, 8
+ store i32 %9, i32* %3, align 16
+ br label %13
+
+; <label>:10 ; preds = %5
+ %11 = load i8** %4, align 8
+ %12 = getelementptr i8* %11, i64 8
+ store i8* %12, i8** %4, align 8
+ br label %13
+
+; <label>:13 ; preds = %10, %8
+ %14 = phi i32 [ %6, %10 ], [ %9, %8 ]
+ %15 = add nsw i32 %.01, 1
+ %16 = icmp sgt i32 %15, 0
+ br i1 %16, label %5, label %._crit_edge
+
+._crit_edge: ; preds = %13, %0
+ %.0.lcssa = phi i32 [ %count, %0 ], [ %15, %13 ]
+ call void @llvm.va_end(i8* %1)
+ ret i32 %.0.lcssa
+}
+
+declare void @llvm.va_start(i8*) nounwind
+
+declare void @llvm.va_end(i8*) nounwind
diff --git a/test/CodeGen/X86/vastart-defs-eflags.ll b/test/CodeGen/X86/vastart-defs-eflags.ll
new file mode 100644
index 0000000000..6017753fc8
--- /dev/null
+++ b/test/CodeGen/X86/vastart-defs-eflags.ll
@@ -0,0 +1,23 @@
+; RUN: llc %s -o - | FileCheck %s
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx10.10.0"
+
+; Check that vastart handling doesn't get between testb and je for the branch.
+define i32 @check_flag(i32 %flags, ...) nounwind {
+entry:
+; CHECK: {{^}} testb $2, %bh
+; CHECK-NOT: test
+; CHECK: {{^}} je
+ %and = and i32 %flags, 512
+ %tobool = icmp eq i32 %and, 0
+ br i1 %tobool, label %if.end, label %if.then
+
+if.then: ; preds = %entry
+ br label %if.end
+
+if.end: ; preds = %entry, %if.then
+ %hasflag = phi i32 [ 1, %if.then ], [ 0, %entry ]
+ ret i32 %hasflag
+}
+